Railway MCP Server

The Railway MCP Server is the official Model Context Protocol server maintained by Railway for interacting with your Railway account from AI agents like Claude, Cursor, and VS Code. It provides a set of opinionated workflows that wrap the Railway CLI, letting agents create projects, deploy templates and services, manage environments, set variables, generate domains, and fetch logs through structured tool calls.

The server can be used in two ways: as a local stdio server via the @railway/mcp-server npm package (which shells out to your authenticated Railway CLI), or as a remote hosted server at https://mcp.railway.com installed via railway mcp install --remote. Either way, authentication is handled through the Railway CLI, so users must be logged into Railway before the tools work.

Destructive operations (like deleting projects or services) are intentionally excluded from the server by design, making it safer to give AI agents access. The server is open source under the official railwayapp GitHub organization and is the canonical way to give AI assistants access to Railway infrastructure.

Tools

Prerequisites

• Railway CLI installed and authenticated (railway login)
• Node.js (for the npx runner)

Quick install

The fastest path uses the Railway CLI helper:

bash <(curl -fsSL cli.new) --agents -y

If you already have the CLI:

railway mcp install

For the hosted remote server at https://mcp.railway.com:

railway mcp install --remote

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "railway-mcp-server": {
      "command": "npx",
      "args": ["-y", "@railway/mcp-server"]
    }
  }
}

VS Code

Add to .vscode/mcp.json:

{
  "servers": {
    "railway-mcp-server": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@railway/mcp-server"]
    }
  }
}

Claude Code

claude mcp add railway-mcp-server -- npx -y @railway/mcp-server

GitHub Copilot

Add to ~/.copilot/mcp-config.json:

{
  "mcpServers": {
    "Railway": {
      "type": "local",
      "command": "npx",
      "args": ["-y", "@railway/mcp-server"],
      "tools": ["*"]
    }
  }
}

Notes

• The server shells out to the Railway CLI, so auth is whatever account is logged in via railway login.
• Some features in get-logs require Railway CLI v4.9.0 or newer.

• Spin up a new Railway project from a repo and deploy a template (e.g. Postgres, Redis) without leaving your editor.
• Have an AI agent fetch deployment logs and diagnose why a build failed.
• Bulk set environment variables across multiple environments when promoting a release.
• Generate a public railway.app domain for a freshly deployed service.
• Audit all projects, services, and environments tied to a Railway account from one chat session.

• "List all my Railway projects and show which services are in the staging environment."
• "Deploy the Postgres template into my current project and link it to the production environment."
• "Set STRIPE_API_KEY and SENTRY_DSN on the api service in the production environment."
• "Pull the last 200 lines of logs for the worker service and tell me why the build is failing."
• "Create a new preview environment for this project and generate a railway.app domain for the web service."

• Officially maintained by Railway under the railwayapp GitHub organization.
• Available as both a local stdio server and a hosted remote server at mcp.railway.com.
• Destructive operations are intentionally excluded, reducing risk of agents wiping infrastructure.
• One-command install paths for Cursor, VS Code, Claude Code, and GitHub Copilot.

• Requires the Railway CLI to be installed and authenticated locally; not purely API-based.
• No destructive actions means deletes and teardown must be done manually outside the agent.
• Some log features depend on Railway CLI v4.9.0 or newer.

• Render MCP server for similar deploy-platform workflows on Render.
• Fly.io community MCP servers for Fly app management via AI agents.
• Vercel MCP for managing Vercel projects and deployments.